Conclusion

This chapter made the word model precise.

A model is a parameterized function. It receives input, uses stored parameters, and computes a prediction. During one forward pass, the parameters do not change. During training, later steps use errors, losses, and gradients to adjust those parameters.

The chapter also introduced capacity as a function-family idea. A model architecture defines what kinds of functions are possible. The parameters choose one member of that family.

The next chapter will make the first useful model family concrete: linear models and score vectors.